SeaDronesSee (SeaDronesSee: A Maritime Benchmark for Detecting Humans in Open Water)

SeaDronesSee 是一个大型数据集，旨在帮助开发在海上场景中使用无人机 (UAV) 进行搜索和救援 (SAR) 的系统。构建有助于 SAR 任务的高度复杂的自主无人机系统需要强大的计算机视觉算法来检测和跟踪感兴趣的对象或人员。该数据集提供了三组跟踪：目标检测、单目标跟踪和多目标跟踪。每个赛道都包含自己的数据集和排行榜。目标检测：5,630 个训练图像、859 个验证图像、1,796 个测试图像 单目标跟踪：58 个训练视频剪辑、70 个验证视频剪辑和 80 个测试视频剪辑光谱镜头：多光谱目标检测：246 张训练图像、61 张验证图像、125 张测试图像 我们将继续更新此数据集，使其更加通用并反映动态情况下的实际需求。

SeaDronesSee is a large-scale dataset developed to assist the development of unmanned aerial vehicle (UAV)-based search and rescue (SAR) systems in maritime scenarios. Building highly sophisticated autonomous UAV systems that support SAR missions demands robust computer vision algorithms for detecting and tracking objects or personnel of interest. This dataset includes three benchmark tracks: object detection, single-object tracking, and multi-object tracking. Each track has its own dedicated dataset and leaderboard. Object detection: 5,630 training images, 859 validation images, and 1,796 test images. Single-object tracking: 58 training video clips, 70 validation video clips, and 80 test video clips. Multispectral object detection (spectral lens modality): 246 training images, 61 validation images, and 125 test images. We will continue to update this dataset to enhance its generalizability and reflect practical requirements in dynamic real-world scenarios.

搜集汇总

背景与挑战

背景概述

SeaDronesSee是一个专注于海上无人机搜索救援场景的大型计算机视觉数据集，旨在通过提供目标检测、单目标跟踪和多目标跟踪三个任务赛道的数据，支持开发高效的自主无人机系统。数据集包含丰富的图像和视频剪辑资源，例如目标检测赛道有5,630个训练图像，并涵盖多光谱数据，以反映实际动态需求。该数据集由图宾根大学于2022年发布，为海事环境下的人员检测和跟踪提供了标准化基准。